Instead of drowning in its own notes, the agent actively curates them. Every turn it asks: what’s worth keeping, and what can I compress? It runs a loop that looks almost human think → fold → explain → act and builds its own hierarchy of ideas.
GENERATIVE AI
-
ChatGPT Full Assistant System Prompt Revealed
By
–
My ChatGPT Pulse this morning titled "Full assistant system prompt"
-
Recursive Language Models for Near-Infinite Context Agents
By
–
One of our best talks yet. Thanks @a1zhang for the amazing presentation + Q&A on Recursive Language Models! If you're interested in how we can get agents to handle near-infinite contexts, this one is a must. Watch the recording here!
-
Using ChatGPT for Calculations via Code
By
–
If you want ChatGPT to calculate for you, use this: "[INSERT WHAT YOU NEED TO CALCULATE] Use code."
-
Creative AI Workflow Using Multiple Generative Tools
By
–
tools:
Midjourney
Hailuo 2.0 (99% of shots)
Kling (opening shot)
Adobe Firefly
Magnific
Enhancor
Elevenlabs source: https://
reddit.com/r/ChatGPT/s/FE
1EY6UlmV
… -
The Dependence of LLMs on Wikipedia Data
By
–
LLMs would be really dumb without the existence of Wikipedia.
-
Cognition SWE-1.5 Speed Optimizations Reach 1881 Tokens
By
–
Just deployed additional speed optimizations for @cognition SWE-1.5. The fastest measured request was an eye watering 1,881 token/s per Grafana dashboard.
-
Google Labs launches Pomelli AI marketing tool for SMBs
By
–
No tricks here, only treats . Here’s this week’s roundup: — Our latest experiment from @GoogleLabs
, Pomelli, is an AI marketing tool designed to help SMBs connect with their audiences faster
— You can now generate a full set of exportable slides in the @GeminiApp — We added a -
Frontier Models vs GPT Codex: Speed to Success Comparison
By
–
Even frontier models fail all the time. The difference is we fail in 15 sec and with one more prompt you get the right answer. Time to success = 30 sec. On GPT Codex it takes 22min just to find out it failed.